A Practical Procedure for the Construction and Reliability Analysis of Fixed-Length Tests with Randomly Drawn Test Items

Abstract

A procedure to construct valid and fair fixed-length tests with randomly drawn items from an item bank is described. The procedure provides guidelines for the set-up of a typical achievement test with regard to the number of items in the bank and the number of items for each position in a test. Further, a procedure is proposed to calculate the relative difficulty for individual tests and to correct the obtained score for each student based on the mean difficulty for all students and the particular test of a student. Also, two procedures are proposed for the problem to calculate the reliability of tests with randomly drawn items. The procedures use specific interpretations of regularly used methods to calculate Cronbach’s alpha and KR20 and the Spearman-Brown prediction formula. A simulation with R is presented to illustrate the accuracy of the calculation procedures and the effects on pass-fail decisions.